Term
Auto-Classifier
A safety layer in Claude Code: a small LLM evaluates every tool call that is not explicitly allowlisted, live, and blocks destructive or unauthorised actions before they execute.
Auto-Classifier — explained in more detail
The Auto-Classifier is an additional safety layer in Claude Code. It sits between the acting model and the tools that actually run — between decision and execution. Every tool call that is not explicitly cleared by an allowlist is judged live by a small, separate LLM and can be blocked before it has any effect.
How it works
For each tool call the classifier inspects context, arguments and target location. Calls flagged as critical typically include violations of a defined state machine, unlisted writes to files or databases and destructive actions such as rm -rf, DROP TABLE or force pushes. Those calls are blocked or escalated back to the user.
Distinction
The Auto-Classifier is not a database check and not code inside the tool itself. It is an extra, LLM-based safety layer above the regular permission configuration — conservative by design: better to block once too often than to let a destructive action through.
Entdecke mehr
Saving Tokens with Claude: 6 Principles That Make Experts Twice as Fast
How I turned my CLAUDE.md from a style guide into a token budget — 6 principles for lower cost, less waiting, and more honest reporting.
GlossarPermission Mode
A global setting in Claude Code that controls how tool calls are handled: Plan (planning only), Default (asks for confirmation), Accept Edits (files free), Bypass (allow everything).
LexikonLangGraph
LangGraph as the standard for agent orchestration — nodes, edges, state, loops, human-in-the-loop and persistence explained clearly.